A different kind of weapon focus: simulated training with ballistic weapons reduces change blindness

Method

Materials

All stimuli were taken from Link’s Crossbow Training for the Nintendo Wii. During training, participants played through six pre-selected levels that introduced the player to a wide range of environments, agents, and destructible objects. The Wii console uses an infrared sensor placed near the display to detect infrared light emitted by the controller. Consequently, it tracks movement of the controller, allowing the player to interact with stimuli on the display with high spatial and temporal resolution. In this experiment, the controller is inserted into the Wii Zapper, which is gun-shaped with a trigger and is held in a two-handed posture like a rifle (Fig. 1). The player plays by pointing and shooting at enemies and objects on the display. After a fixed amount of time, the perspective moves. Because the game is of the rail shooter genre, the player’s perspective is fixed, but they have the freedom to look around, aim, and shoot. Thus, differences in visual experience between participants in the training phase can be attributed entirely to which things they do or do not shoot and where they direct the reticle of their weapon. Critically, participants all saw the same agents, objects, and environments.

At test, participants viewed images from the game they had just played. These images were screen-grabbed and altered using Photoshop. The alterations were always the presence or absence of an element: either an agent, object, or a feature of the environment. There were ten unique image sets for each category, for a total of 30 test stimuli. Each type of change was presented equally often in different settings. For the forest environment, for example, the change was equally likely to be an agent, object, or environment. All participants saw the same 30 image sets.

The training game and the stimuli for the test phase were displayed on a projector facing a white wall in a dark room. Participants stood behind the projector, where they completed both the training and test phase.

Procedure

During training, all participants played six levels of the game, each lasting between 1 and 2 min. The duration of each level was fixed, controlling for exposure to the training stimulus. Participants stood behind the projector, using the weapon to point and shoot at items on the display. In three of the six levels, the task was to shoot objects such as barrels or targets. In the other three levels, their task was to shoot threatening agents that would approach them on the display. This ensured that participants had experience shooting agents and objects.

Participants were assigned to one of two search groups for the test phase: holding the same weapon with which they had trained, or a hand-sized ball. Participants were assigned to each group in alternating order. Participants initiated a trial by pressing and holding a large button on the table in front of them with their object (weapon or ball). After 1000 ms, a source image and a modified image were presented in alternating order for 250 ms with interleaved gray masks (Fig. 2). Participants searched the image for a change or until 40 s had passed, at which point the trial was terminated. Upon detecting a change, they raised their tool and pointed it at the changed item. Releasing the button ended the trial, measuring reaction time (RT), and ending stimulus presentation. Participants were then required to name the change. The experimenter input the response on a separate monitor not visible to the participant. No feedback was given. Participants knew the change would occur to an agent, object, or environment and that the change would be the alternating presence and absence of an item. Participants were instructed to react as quickly as possible while minimizing errors. The 30 flicker stimuli were presented in a randomized order. The experiment was 2 (search: weapon or ball; between-subjects) × 3 (change: agent, object, or environment; within-subjects) mixed-factors design.

Results and discussion

Responses for which the participant failed to identify the change before the trial timed out (40 s) or incorrectly identified the change were categorized as errors. Mean RTs for correctly-identified trials for each participant were entered into a 2 (search tool: weapon or ball; between-subjects) × 3 (change stimulus: agent, object, or environment) ANOVA. One participant’s data could not be entered into this ANOVA because they had no successful detections in one cell and thus was excluded from further analyses. Change type significantly influenced RT, F(2,34) = 49.58, p < 0.001, η_p ² = 0.74 (Fig. 3). Participants were significantly faster in detecting changes to agents than to objects or the environment (ts > 3.65, ps < 0.002), replicating the finding that attention prioritizes animate things (New, Cosmides, & Tooby, 2007).¹ Importantly, for the purpose of our study, there was no effect of search tool and no interaction, Fs < 0.11, ps > 0.897.

The mean number of errors was examined with the same ANOVA. Errors include misses and false alarms. There was a significant main effect of change type, F(2,36) = 22.38, p < 0.001, η_p ² = 0.55. Because participants were both faster and more accurate for changes to animate objects, we can discount the possibility of a speed-accuracy trade-off. No other effects reached significance, Fs < 1.20, ps > 0.314.

We expected that participants would rapidly adapt to the actions afforded by the weapon and that holding that weapon during search would increase the likelihood of detecting shootable things. Instead, we found that it made no difference whether they searched holding the weapon or an inert ball. It is possible that the training phase did alter their allocation of attention and that they simply simulated those actions in both the weapon and ball conditions during the test phase. Alternately, it is possible that training with a weapon had no effect on attention during search, where the weapon was inert. In other words, weapon training may only change the allocation of attention while holding a weapon that is “on.” To assess these possibilities, we ran a second experiment, with fully crossed training and test phases.

Method

Results and discussion

Responses for which the participant failed to identify the change before the trial timed out (40 s) or incorrectly identified the change were categorized as errors. One participant was removed prior to analysis for being hungover. Mean RTs for correctly identified trials for each participant were entered into a 2 (training: weapon or ball; between-subjects) × 2 (search: weapon or ball; between-subjects) × 3 (change: agent, object, or environment) ANOVA. There was a main effect of change type, F(2,86) = 158.24, p < 0.001, η_p ² = 0.79 (Fig. 4), replicating the agent-specific benefit over objects and environments (ts > 6.75, ps < 0.001). Critically, there was a main effect of training tool, F(1,43) = 7.35, p = 0.010, η_p ² = 0.15. Search times were faster for those who had trained with the gun, actively interacting with the display, compared with those who had searched while holding a ball. Moreover, the interaction between training tool and search tool was approaching significance, F(1,43) = 3.43, p = 0.071, η_p ² = 0.07. No other main effects or interactions reached significance, Fs < 1.14, ps > 0.292.

The mean number of errors was examined with the same ANOVA. There was a significant main effect of change type, F(2,86) = 28.71, p < 0.001, η_p ² = 0.40, again showing that there was not a speed-accuracy trade-off. There was an interaction between change type and training tool, F(2,86) = 3.63, p = 0.03, η_p ² = 0.08, which was qualified by a three-way interaction, F(2,86) = 3.67, p = 0.03, η_p ² = 0.08, which is caused by more errors to environmental changes in the ball–ball condition. We suspect this interaction is largely spurious, given the small number of errors overall.

Participants who played the game with an active weapon were more sensitive to changes during a later search even though neither search tool (weapon nor ball) was actually functional during search. One explanation for this finding is that the ability to destroy elements during training was later simulated during search, regardless of which tool was being held. This weapon simulation enhanced the allocation of attention across the entire display. Interestingly, this effect did not interact with change type, suggesting that the attentional improvement was uniform rather than specific the destructible items. And although it should be interpreted with caution due to its marginal significance, there was some evidence of an interaction between training tool and search tool, as participants who watched the videos and searched with the ball were slower than groups who had experience with the weapon.

Method

Results and discussion

Responses for which the participant failed to identify the change before the trial timed out (40 s) or incorrectly identified the change were categorized as errors. Unlike Experiments 1 and 2, the training and search factors were not fully factorial, so treating the three conditions as a between-subjects factor is appropriate. Moreover, the lack of training and search interactions with change type motivated us to focus on the comparison between the weapon–weapon condition and the two control conditions.² Consequently, mean RTs for correctly identified changes were entered into a between-subjects ANOVA contrast between the weapon–weapon condition and the other two conditions, where we observed a significant effect, F(1,31) = 5.79, p = 0.022, confirming that participants in the weapon–weapon condition detected changes faster than the controls (Fig. 5). The same contrast was conducted on the mean number of errors; there was no effect (F < 1).

Participants who played the game detected changes faster than participants who held onto the simulated weapon and watched the game or participants who played the game with a controller that used a non-weapon posture. These results resolve the confounding of training object and training task in Experiment 2 and, in addition to replicating our result, they rule out the alternative explanation that the present training effects are caused by general arousal or engagement with the game.

General discussion

Across three experiments, participants trained and searched for changes while holding a weapon or a ball. During training, participants using the simulated weapon could actively destroy on-screen elements, whereas participants using the ball could not. We found that participants who had trained with the weapon were later better able to detect changes to stimuli regardless of the object that changed (agent, object, or feature of the environment). This result shows that vigilant surveillance benefits from sensorimotor training with the tool – in this case, a simulated weapon – that you would use to interact with the environment.

Another important finding is the null effect of the search tool. It made no difference whether observers searched with the weapon or ball. This violated our expectations, given that action affordances flexibly alter attentional allocation during change detection (Symes et al., 2008) and that tool use in general – and guns specifically – alter attentional location (Biggs et al., 2013). Accordingly, we predicted that holding the weapon during search would confer advantages, or at least produce biases, compared with holding the ball.

We can, however, cautiously interpret a marginal interaction between the training tool and the search tool in Experiment 2. Participants who trained and searched with the ball were slowest to detect changes across the board, whereas participants who trained with the ball (no sensorimotor training; only visual) and searched with the gun expressed search latencies more like the groups who trained with the gun. It appears that any exposure to the weapon, whether during training or search, was sufficient to attenuate change blindness. The tantalizing implication drawn from this result is that participants who trained with the ball but searched with the weapon were as capable of simulating the weapon-related action affordances during search as those who trained with the weapon.

Interestingly, experience with the weapon led to an item-non-specific attenuation of change blindness. In other words, changes of all types were detected faster after training with the weapon. If the observed attenuation of change blindness is indeed caused by variations in attentional allocation resulting from simulated weapon affordances, then we might expect that shootable elements – agents and objects – should be prioritized during search over environmental elements in the weapon conditions. Instead, we found that simulated weapon training led to a blanket improvement in change detection times regardless of change type, suggesting that change detection is universally improved by sensorimotor training. This finding sets the current study apart from existing research on domain-specific expertise and change blindness. Experts in solving physics problems (Feil & Mestre, 2010) or watching football (Werner & Thies, 2000) are faster at detecting changes to domain-relevant images as compared to novices. However, these studies examine true expertise derived from perceptual experience that goes well beyond the ~10 min practiced by our participants. Moreover, the item-non-specific attenuation of change blindness we observed sets the present findings apart from domain-relevant attenuation of change blindness. A domain-relevant effect in the present study would occur only for shootable things (agents and objects). We found no interaction between training tool and change category, indicating that the attenuation of change blindness was not domain-relevant as it is in existing investigations of perceptual expertise.

Additionally, the effect of change type, wherein changes to agents were detected faster than inanimate elements, draws comparisons to the animate monitoring hypothesis (New et al., 2007). It is not surprising that agents should be detected fastest, especially considering that they were all threatening stimuli in the present study, however it should be noted that destructible objects were detected much faster than environmental elements. In previous studies examining animacy and change detection, artifacts and topographical landmarks are detected equally quickly (New et al., 2007). In contrast, we found a large advantage for objects over environments. We propose that objects’ affordance for destruction led to improved attentional allocation over the environmental changes.

In many real-world scenarios, the critical change – for example, the sudden appearance of a threat – occurs only once, rather than alternating as in the flicker paradigm. For this reason, it would be helpful to conduct future investigations using the one-shot change detection paradigm, where an image is presented for a short duration, followed by an intervening mask and a slightly changed image (e.g. Phillips, 1974). We used the flicker paradigm, where the images alternate repeatedly, because we reasoned the potential for weapon use would occur in scenarios where the observer was engaged in prolonged surveillance rather than a brief exposure. Both methods provide unique insights into the nature of change detection (Rensink, 2002) and the applied question of change detection during vigilant surveillance. For example, the flicker paradigm is better suited for measuring the speed of change detection, whereas the one-shot paradigm typically measures accuracy.

Experiment 3 was conducted with the intent of isolating the best control conditions to compare against the weapon–weapon training-search condition. Because of these comparisons, we can confidently attribute our finding to sensorimotor training with a simulated weapon: change blindness was attenuated following simulated training with a weapon compared to visual-only and arousal-matched control conditions. Another interesting comparison would have been a no-training condition, where participants had no exposure to the stimuli at all prior to the change detection task. The marginal interaction between training and search conditions in Experiment 2 suggests it is possible that sensorimotor training is not required to attain this improved change detection; searching with the simulated weapon may be sufficient. Consequently, it may be possible that an observer with no prior exposure at all to the stimuli could achieve a similar improvement in change detection just by holding a weapon. This speculative result imagines a strong weapon affordance effect on attention.

In the real world, training regimens for police and military duties extend far beyond the ~10 min practiced by our participants. Because the training period was short, and because there was no delay between training and search, we can only speak of short-term effects on vision. The effect of simulated weapon training can be attributed to sensorimotor experience, but it is possible that long-term training effects can be elicited with only sensory training. More research is necessary to determine whether sensory-only training can produce long-term improvements in attentional allocation in sentry-type tasks. The restriction of our conclusions to the short term reveals an additional point of interest: that we observed attenuated change blindness as a result of simulated weapon training indicates that these types of training affordances incur immediate effects on attentional allocation, consistent with a rapid integration of tool affordances into the body schema. This is concordant with some of our earlier research showing that brief exposure to a new tool stimulus is sufficient to alter the allocation of attention to and near that tool (Taylor & Witt, 2014, Experiment 4; also Reed et al., 2010).

In conclusion, training with a simulated weapon led to immediate item-non-specific attenuation of change blindness. This effect reveals the importance of training regimens when it comes to the use of weapons and more generally suggests that sensorimotor experience with a tool is essential to activate action-related attentional priorities.